Revolutionizing Education with AI | Khan Academy founder Salman Khan

26:54So even though historically, many people associated teaching with, oh, I'm going to write a lesson plan, then I'm going to deliver this lecture, then we're going to have a quiz every two weeks, and then I'm going to grade those quizzes, and then we're going to rinse and repeat for two weeks, any great teacher will tell you that's actually not their job. What's actually their job is, on a very human level, to connect with their students, either individually, in small groups, as a whole class, to motivate them to see the wonder in the world, to make sure that these students feel seen and supported.

27:33And then on top of that, the teachers have to grade papers, write lesson plans, progress reports, IEPs, you know, prepare them for standardized tests, etc. And because we're asking so much of teachers, and some of that latter half of stuff is more tangible and more measurable, it's actually squeezed out a lot of teachers' opportunities to do those very human things. I think anyone who goes into the teaching profession, they dream of these moments where they're able to have some time in a small group with some students or give that one student who was really down and out a pep talk and it changes their life forever.

28:10And it will change their life forever. I think, you know, in my 12 years, 13 years in the K-12 system, I can, I remember probably every moment when a teacher did single me out out of the 30 kids and say, hey, Sal, can I talk to you about something? I really liked what you did, or I really didn't like what you did. That, it affected me. It changed who I am. And I think that's what, or a teacher ran a Socratic conversation with 10 of us, as opposed to all 30 at the same time, or we had a field trip where we ran a simulation. And those are the things that I don't think the AI is going to be able to drive, but I think the AI will be able to support it.

28:46And, but once again, freeing up the teacher to do, to do more of that. The jobs that I, I think will be under threat with AI, or there will be fewer of them, are the ones that you are you know, if great writers should not feel threatened, but if you're the person who writes the fairly generic text every day on the news sites about why the stock market went up or down, I'm surprised if those aren't already written by AIs. If... I think a lot of them actually are. They feel like it. They feel like they've been written by AIs for 10 years. Yeah. I think if you are, you know, if I'm thinking of the education system, the registrar's office, I think, is going to be, you're going to see a lot of automation there.

29:33But once again, as just a citizen of the world, I think the more resources that we can save on the non-student-facing thing and put it onto the student-facing thing, that's a good thing. If you look at school districts like New York City, they spend an average of$40 ,000 per student per year. There are 25 students in an average classroom in the New York City Department of Education. So that means if you just took$40 ,000 and multiplied it by those 25 students, that means that there's a million dollars. I guarantee you that no matter how good the teacher's benefits, pensions, etc., they're getting a small fraction of that million dollars.

30:16Maybe a very senior teacher with a great pension, etc., maybe with great health insurance, maybe it's$150 ,000 if you put all the benefits in there. you know, approaching 200 ,000. That's very, I think I'm already being overly, you know, no one's really making that much. So where are all the other resources going? It's actually not going to the real estate in most cases. Where is it going? And it's going to all of this, let's call it back office, administrative stuff that really isn't moving the dial with students. And actually, I don't write too much about that aspect of it. That's probably less interesting for a lot of folks.

30:52But yeah, registrar's office scheduling, you know, just a lot of the administrative things, I think will hopefully get much more streamlined. You know, what about the current education system is an antique at this point? I mean, I feel like the current, roughly the current education system's been around for 200 or so years. That you go to a school, you go to a location where there's a bunch of subjects, taught throughout the day. So every day it's like six different subjects. You go from class to class to class and you get homework at the end, you get tested. So that's roughly the structure of the modern education system.

31:34And what part of that do you think is outdated now? Yeah. And, you know, my first book that I had written back in 2012, One World Schoolhouse, it kind of goes into how did we end up with the system? And it really does, it's not a coincidence when you say it's roughly 200, 250 years old, it came out of the Industrial Revolution. And for the most part, it was a very utopian idea. For most of human history, we learned through apprenticeship. We learned through following around, learning from our cousins or our parents how to hunt or how to cook. Once society became more advanced and more specialized, we would hang out with the blacksmith or we would hang out with the, I don't know, the carriage repair guy, whatever it would be to learn.

32:22Or even things like law and medicine was an apprenticeship until about 200, 250 years ago. But the issue with a lot of these is that they didn't scale. If you go back 300 years ago, even in more literate parts of the world, you only had a 30 or 40 % literacy rates. Most of the world, you only had about a 10 or 15 % literacy rate. And so when we had the Industrial Revolution, we had a more abundant society because of technology. And a lot of these societies, and it's not a coincidence that these were some of the first societies to develop a middle class, the UK, what would eventually become Germany, Japan, the United States, they were the first to say, hey, let's have a mass public education system.

33:07Very utopian idea. But they said the only way we can afford to do that, the only way we can scale it is by leveraging the techniques of the Industrial Revolution. We're going to batch students together, usually by age, move them together to set process. Literally a bell will ring every hour. I mean, that directly comes from a factory. We're going to have standards. We now take standards for granted in education, but this was a thing that came out of mass production in a factory. We will assess periodically. and, but we're not going to slow down that assembly line. Instead, we're going to assess and some of the quote product is going to be, you know, said, okay, this is the product that's going to be the doctors, lawyers, engineers, and some of it, well, the information or the knowledge doesn't seem to be sticking.

33:54Well, these will be the less skilled laborers. So it kind of worked. You know, we could talk about, was it fair for folks? You know, you could have been, And people have been talking about the inequities of tracking for decades. You know, when you're 12, something's not sticking. And all of a sudden, you're tracked into a slower track. And now, all of a sudden, no one expects you to go to college. And people expect you to, you know, be a lower skilled laborer versus becoming a doctor or lawyer. Is that fair? That's a very big question. I don't think it is. But it kind of worked for society in that we didn't need a huge number of people in the knowledge economy.

34:31I mean, we did need a lot of people who are in either the middle income or lower income jobs, a lot of the mid-skilled or lower skilled labor. We did need it. But if you imagine the world that we are going into, we know what's going to happen to the less skilled jobs, you know, or even the mid-skilled jobs. Obviously, we're talking about artificial intelligence, which is all about kind of that mid-skill processing of information. Everyone I talk to, robotics is not far behind. I think a lot of folks are saying in the next five or 10 years, we're going to have a chat GPT type moment with robotics as well.

35:07So that also means that some of the lower or mid-skilled labor is also going to be automated probably in the next 10 or 20 years. And so where does that leave us? Well, it leaves the high-skilled labor, the knowledge economy, being a researcher, being an engineer, being an entrepreneur is where all of the productivity, all of the wealth will accrue there. And so as a society, we have to decide. Are we okay with only right now 10 % of people participate there? And then that's not a stable society. Are we going to have to do redistribution, which to me is still pretty dystopian because people want to have a sense of purpose?

35:41Or do we leverage the technologies that we have to get as many people as possible into the top of that pyramid to be in the knowledge economy? Instead of having a labor pyramid, you can have an inverted pyramid where most people are able to be in that knowledge economy. And so I think that's where the imperative is. So how do you start making those changes in the educational system? And also, I think there's some question as to whether people effectively learn by taking six different subjects in a day. Like multitasking learning has been shown to be not the most effective way to learn. And you even mentioned this in the book, that the more one-on-one, more immersion is probably better for education.

36:20Yeah. You know, the core ideas, and this is something I've been preaching well before artificial intelligence came on the scene is instead of a traditional model, what we hold fixed is how long you get to work on something because that assembly line keeps moving. And what's variable is how well you learn it. And we get that in the form of grades. You know, you and I take a class together. We're in algebra for this term. You got an A minus. I got a C plus. All right. It goes in my permanent transcript and we move on. Somehow expecting me that I got to, you know, expecting someone who got the C plus to now understand algebra two or understand something more advanced.

36:55And so what I've been advocating for the last 10 years or more is, let's do it the other way around. What should be variable is when and how long you work on something, and what's fixed is how well you learn it. And that's what a personal tutor, when Aristotle was Alexander the Great's tutor, I'm sure that's what he did. If young Alexander hadn't mastered a concept yet, he would have slowed down a bit or given him a second chance to take that test to make sure that he really mastered whatever military strategy or, you know, governance, whatever was the topic for the day. Now, if I said this, say, 30 years ago, people would have rolled their eyes.

37:37They'd say, well, okay, it's easy for you to say, but unless you have a lot of resources to provide that level of personalization, there's no way you can have 30 different students learning at 30 different paces. and if one student is falling a little bit behind, for them to have a second chance to take that test, that's just logistically impossible to do. Now, what's changed over the last 30 years, even before artificial intelligence, is you have tools like Khan Academy. You have software that can let every student practice at their own time and pace, give teachers real-time information. If a student needs to refresh their knowledge on something, they could watch an on-demand video, they could read an article.

38:15Now they can have a conversation with an artificially intelligent tutor. And once again, that tutor is also in contact with the teacher. So constantly guiding the teacher on how they can be a conductor of this class of 30, where they don't have to make every person play the same thing at the same time, regardless of how bad the quality or how not ready for it they are. Now you can personalize a lot more. And so, and a corollary to that is, you know, I talk about mastery learning, which is if you haven't learned it well yet, keep trying. And once again, this is how we've always and still do. If you want to be an Olympic athlete, if you want to be a musician, you are doing mastery learning.

38:57We just don't do it in our school system. And you see the outcome differences between how good people can get as if they have a personal coach. But a corollary to mastery learning is what I would call competency-based learning. Right now, our entire credentialing system is based on seat time for the most part. Did you sit in a chair for 12 years and kind of do what you were told, even though you don't really understand a lot of it? Okay, we'll give you something called a high school diploma. Did you sit in a chair for 12 hours a week for a term? Okay, we'll give you the 12 credit hours or however they want to account for it for that course.

39:31And you kind of learned it. We know in reality, very few people actually retain most of what they quote are exposed to in school. 60, 70 % of kids going to community colleges have to get remediation, not even at a high school level, essentially at a seventh grade level. So they sat in these classes called Algebra 1, Algebra 2, Trigonometry, some of them even take Calculus. And the colleges are saying, you're not even ready to learn algebra yet. Go back, we're going to work on your pre-algebra. So a corollary to personalization and mastery learning is moving to a competency-based credentialing system.

40:03If you know it, take some type of a rich assessment and we'll give you credit, even if it only took you two days to learn it. But if you haven't learned it yet, that's cool too. Here are some resources to keep learning. Come back in a month, come back in two months, come back in a year, and I'll give you a second chance to actually learn the material.

40:36So now let's bring in AI. How would you use AI to, who would decide, would it be the student or the teacher or both or the AI who would decide like, oh, James needs a little more work on his algebra two, as opposed to Sal who got an A plus on every test, who would decide how I, you know, how much more time do I need? And how, what would I be assigned to do? Like, would the AI give me tests or teach me or like what would happen? It's a little bit of all of the above where even before we had artificial intelligence, we have these classrooms where students are working on Khan Academy and we were able to give signals to teachers as to which students are moving ahead and we want to encourage them to do so, which students are on track and which students are maybe falling behind.

41:21But we did leave it all, you know, we did professional development for teachers on how they might want to navigate that. Hey, for the kids who are ready to move ahead, let them move ahead. For the kids who are behind, why don't you do a focused intervention with those six or seven kids? And a lot of teachers were doing that type of thing, but you can imagine it can get pretty complex and pretty hard, even if you have some of these software tools. What we're able to do now with the artificial intelligence is it really is acting like a data analyst and or teaching assistant for these teachers. So instead of a teacher every night having to look at a spreadsheet like dashboard and say, okay, these are the kids who are struggling.

41:56Let me break them out. Let me do another lesson plan, which is a lot of work for the teacher. They can just ask the AI what's going on. And the AI says, look, here's what's going on. These are the kids on track here. You We do a differentiated lesson plan tomorrow. I recommend these kids work on this task while you spend the first 20 minutes with these students. And the teacher's in charge. This thing is in support of the teacher. The teacher is going to be able to say, no, I don't really like that activity. Can you? And we already have this where they can highlight parts of it and they can say, let's come up with something that's a little bit more fun or a little bit more engaging or gets the kids out of their chair.

42:31And the AI will say, yes, sir. Yes, ma 'am. And do that. But as you can imagine, it dramatically lowers the amount of analysis and prep time that a teacher has to spend in order to create these really engaging and differentiated personalized education experiences. It's also going to all these other trappings of school that as students we never saw, as parents we oftentimes don't see. Writing progress reports, grading papers, IEPs, which are these plans you have to write for students who need special supports, which are increasingly a larger and larger percentage of the student population, takes a lot of the teacher's time.

43:13Teachers have to prep their own knowledge. If the AI can dramatically improve that time, we've been getting signals from school districts that Conmigo is helping the teachers with this non-student facing work. It's saving them five, 10 hours a week. And so you're going to see it there. You're going to see there's this phenomenon called learning management systems, which are increasingly the ways that teachers and students communicate with each other and make assignments and submit assignments. And right now, these are web-based things. I see more and more the AI acts as that intermediary where it's working with the teacher on the planning, on the creation.

43:52It provisions it to the student. It works with the students on those things. And then it's able to report back to the teacher. So, you know, and I can daydream more and more. You can imagine a world where an AI can start to even know what's going on in a classroom and support even more. But even before we go into those more sci-fi use cases, I think it's just going to be in the fabric of everything that the teacher and the students do and hopefully in a way that supports the students and the teachers better and streamlines things. Do you think it will happen? Like, let's say, you know, AI is moving pretty fast and already there are tools like Conmingo just, you know, what is it, like a year and a half after the first ChatGPT was released.

44:32Where do you see education five years from now? like in K through 12, and then I'll ask about other types of education. Yeah, you know, I think there are certain trends that are going to happen that are not fully technologically related, and there's going to be things that are going to be AI related. The non-technological related, I hope we do move, as I mentioned earlier, to a competency-based system, less seat time, that we can go more to a mastery-based system. And that starts to involve a little bit more technology, because if you're asking a teacher to say, hey, if a student got to see the first time, they should have another chance at that assessment.

45:03Well, who's going to write that assessment, that's work. Who's going to grade that assessment? That's work. And so I think that's where the technology plays into it. In terms of how the technology is going to manifest itself, for sure, all of that support work that we used to bury teachers with, I think is going to be dramatically streamlined. Once again, lesson planning, writing exit tickets, grading papers. Teacher's always in charge, but I mean, imagine being a seventh grade English teacher and every two weeks you literally have to sit there and read 100, 150 papers about grade expectations from seventh graders.

45:45I could imagine by the third or fourth paper, it can get a little bit tedious. Instead, you had a reliable AI that can give analysis. You can spot check it. You're the final arbiter, but it could take that thing that was ruining your weekend to a task that might take an hour instead of 10. So you're going to see that. You're going to see the AIs get better and better at supporting the students, not just conceptually, not just academically, but I think from an executive functioning, from a metacognitive. So one project we're working on, we call it Proactive Conmigo. It doesn't just act as a Socratic tutor.

46:22It's going to start messaging students. Like, where are you? The things I used to do with Nadia back in the day. Hey, you said you were going to hit these goals by Friday. It's Saturday now, and you're only halfway. Come on, what's going on? It's going to be able to facilitate a lot more communication between the teacher, the student, and the parent. When it finds that, hey, there's something off, parent, you should know about this. And hey, teacher, by the way, I already told the parent that this is happening. I think you're going to see more of that. I think the whole area of assessment is going to be really, really interesting.

46:55A lot of the criticism of education over the last 20 or 30 years is we've tried to measure more things, which for the most part is a good thing. And we've been trying to measure it in standardized ways, which for the most part is a good thing. But because so many people started indexing on these things that you can measure in a standardized way, they maybe have lost sight on other things. I was talking to a senior administrator at Harvard. And he was telling me that even at Harvard, they're seeing almost an epidemic of kids who can't write. And if Harvard's seeing that, I can guarantee you it's 10 times worse pretty much everywhere else.

47:34And it's probably, you know, no one knows exactly why. Well, think about what standardized testing has been for the last 20 years. There's no writing in it because writing has fundamentally been hard to assess or to be able to give someone standardized clear feedback on. And so it's kind of maybe started to disappear in many cases. We see in the other things in math, when people focus on a multiple choice assessment only, multiple choice can have value. There's nothing inherently wrong with multiple choice, but it might squeeze out other things if we put too much emphasis on it. And so I think AI with its ability to make sense of more open-ended responses and even visual responses or even video responses, I think it's going to broaden the aperture of how we assess and how we even assess in a standardized way.

48:22And that's going to be good because it's going to allow school to going back to, let's not just focus on what can be assessed or narrowly assessed, but what's important. And now we can hopefully assess that in a broader way. So do you think, is this hopeful or do you think this actually will happen in the educational system? We're working on it. So it's, I'm both. There's definitely a possibility that a lot of folks, well-intentioned folks, have good ideas and either they don't get traction in a system that historically has sometimes been slow to move or it manifests itself in unproductive ways.

49:04That's one of my fears. That's why I tell our team at Khan Academy why our role is important. We're a not-for-profit. We are focused on how does this actually drive impact? There are other players, there may be more sales focused, and they're just happy to make the sale, however it gets used. And that might not lead to the best possible outcomes for teachers and students. But generally speaking, I'm, you know, obviously, I'm, I plan on devoting my life to this, so I wouldn't be doing it if I didn't think that this this was was going to happen and so you know you mentioned earlier that a lot of times students don't retain what they learn i think this is really true it's it's what you learn in history class in ninth grade i would say there's probably a one percent chance you remember it now unless it's something that particularly fascinates you what do you think will change in the system that will increase retention?

50:02I think if we are more explicit about moving to a competency-based system where we say, look, well, one, competency-based system makes you have to get more clear about what you care about. Like, what are we really going to... And look, some of the things that we learned in ninth grade that we forgot, that might be okay because it was less about learning the text of... I don't even, you know, I don't remember even the Eighth Amendment. You know, the text of the Eighth Amendment but it was more about learning to critically analyze things, et cetera. And that might hopefully be a skill that we have retained.

50:36But there probably is content knowledge that we do want. In fact, there for sure is content knowledge that we want people to retain, but we need to get just more explicit about that. And then if we get more explicit about, look, everyone who graduates from high school should, for the rest of their life, be able to solve a simple equation. They should for sure know who Alexander the Great was. They should for sure know how to make sense of a informational text written at at least maybe a seventh or eighth grade level, which is what most texts in our life are actually written at. And then if we can focus on those things, because competency-based learning also focuses you, because right now there's a lot of just hoop jumping and seat time and this or that that could be very idiosyncratic to a particular classroom.

51:30It allows you to focus on those things and make sure those things happen. And then hopefully streamlines from some of the other busy work that doesn't have to happen as much. But do you ever see a point where, okay, this student's really good at math. Oh, he spent one week learning algebra one. That was all he was focused on. So now he's able to move on to algebra two and he goes at his pace and another student goes that actually goes at a slower pace and so you do have for a classroom of 30 students you do have people going at 30 different paces you know and and they're immersed so so that increases retention there's less multitasking from going from biology to history to fit to math or whatever so some of that's already happening we after i wrote my last book we and my my oldest at the time was entering kindergarten we started a school, a con lab school that implements a lot of these.

52:19And I won't claim that we figured it all out. In fact, I'm always pushing the school to be thinking a little bit more about questioning a lot of the assumptions. But, you know, at that school, yeah, you do have a good number of students. And math is probably where we're seeing it the most, where they could easily be three, four, five, six grade levels ahead. There's a lot of students who are probably, you know, at or one or two grade levels ahead. And there's a few who, you know, the faculty is working extra hard to make sure that they're going at least at a reasonable pace, that they're going to at least get to calculus before they leave high school, which in the broader world is already a win because most kids don't even get there.

53:00So we're already seeing that type of reality. But I think we need to see more experiments. What you just mentioned of like being able to focus on certain things at a time, I'm intrigued with that idea. There's universities like Colorado College, or colleges, I should say, that do do that. You take classes, instead of taking four or five classes, you take one at a time, but you do it for two weeks, and that's all you do. And then you go and switch and you do another one. I think that's interesting. And I don't think it's just going to be the con lab schools and the Colorado colleges of the world.

53:35We have partnerships with very mainstream school districts. Newark, New Jersey, we're seeing some really good things. and they're doing a bit of a hybrid where the teachers are using Khan Academy to assign the daily practice. And by the way, the kids have support from Khan Migo, but the district says, hey, look, if you finish your assignment and you did it well and you showed your proficiency, you can keep going. But if you did the assignment, but you only got 30 % of it right, you should have another go at it. And now it's going to be different items because it's sampling from a very deep item bank.

54:06Or maybe you're not ready for that assignment because you're missing some prerequisite skills. How do we get you to get some more practice on that without falling too far behind? This is happening in Newark, New Jersey as we speak. It's interesting. In 1982, I was part of this program where these seventh graders would take the SATs. And if you did well, you were invited to participate in this program at Duke University called TIP, where it had this immersion experiment. And I always remember, finally, like in three weeks, I was able to go through like a couple of years of math. And, and then when I got back to regular school, so I was in a summer program.

54:46And when I got back to regular school, I was able to advance more quickly. And I always thought that was an incredible experiment that they were doing. And then at the time there was only 22 students in the program. Now there's like tens of thousands across various campuses. And I just wonder when kind of the mainstream educational system is going to start adapting this, but it sounds like it is to some extent. It is. And if you look at communities where I live, I live in the middle of Silicon Valley, what you're describing is more the norm than the exception. But yeah, and I think the work is, how do we make that accessible and the norm more broadly?

55:23And that's why most of our work at Khan Academy, most of the places where we are deploying Khan Mego are large scale public school districts. But we're seeing more traction than I would have guessed a couple of years ago. Now, in terms of the existential threat for various industries, one area that has been a big concern, and you mentioned this quite a bit, is the ability to write. Like, this has been a big debate in education. Like you mentioned at Harvard, some students don't even have the ability to write. And then in the creative industry, like screenwriting, I think there is a real question.

56:03question, will AI, will the GPT-7 be able to write an Oscar-winning movie? And I think the conclusion in general and your conclusion is that it won't be, but it'll be a useful tool again for the creative to be even more creative. But I think people are very worried about this. And I will tell you just from conversations I've had with heads of major movie studios, they are looking for solutions other than screenwriters? Yeah, well, I think there's a bunch in this. So on the first level, let's take a, you know, most, writing essentially has two pieces to it. There is the putting words on paper in a structured, hopefully engaging way, grammatical way.

56:55And then there's the, how do you come up with what you're going to write? So if you're a journalist, I would think most of your work should be going out there talking to people, looking through public documents, attending public hearings, being on the scene of the crime to report what's going on. So that's all the work that I don't think AI is going to be able to do anytime soon. And then the journalist takes all of that, and then they write the article and all of that. And so I think AI there very clearly has a role of, well, if it can help the journalists, let's say, take all of their notes that they just got from multiple resources, sources, maybe different recordings, maybe even videos, and then help them get to a first draft pretty fast, maybe with a little bit of prompting.

57:42I think that's a win for that journalist. They're going to be able to spend more time on the information gathering and less time on the wordsmithing. But once again, I don't think it's that you're just going to take all that input and let the AI just pop something out that, okay, let's put it in the New York Times. No, the journalist is then going to tweak it and say, no, that, they're going to act more like an editor and they're going to say, no, that, we're burying the lead. Let's put this quote up front, et cetera, et cetera. Let's tighten it. Let's make it a little bit more punchy. So there's still going to be work and the craft of writing and the ability to recognize good writing is still going to matter.

58:16I think if we go to the movie industry, I'm not sure exactly how this is going to play out. But, and I write about this in the book, I think what is going to happen is AI is going to lower, like I found it ironic that it was the screenwriters who are afraid of AI, that the production companies really want it. I actually think that AI is going to bring the balance of power to the individual, to the screenwriters, and take it away from the gatekeepers who are the production houses. We already saw that before AI with things like YouTube and social media. I mean, you're even seeing this with mainstream news now.

58:56It's, I mean, it's essentially been disrupted already by YouTube and social media. The ability for someone to publish and be discovered without having to go through gatekeepers now is completely transformed. You know, Justin Bieber is a YouTube artifact. I'm a YouTube artifact. Like, you know, people can publish podcasts now. You don't have to make a pitch to some executive at wherever to do that anymore. That's the way the world was 30 years ago, 40 years ago. With AI, I think you're going to see a similar, you're lowering the cost of getting into the game. Today, if I have a great idea for a science fiction movie, and I have a few I dream about one day, not only would I have to write a screenplay, I would have to get it in front of people who take me seriously.

59:47you know, even if I wrote a great screenplay, I might not even be noticed unless I'm connected in the right ways, unless I get the right people to read my screenplay. And then someone will say, oh, it's a pretty good screenplay. I'll pay a couple of hundred grand for it. If it's really good, most cases I'll pay a couple of tens of grand for it. You know, go write another one while you're, you know, living off of ramen. And, and, and then that movie studio might throw some of them away, but eventually say, okay, we're going to put a hundred million dollars behind this one, hire actors, director, blah, blah, blah, blah, blah, blah.

1:00:16Five, 10 years later, the movie comes out. If it's a blockbuster, you know, the movie studio is going to make hundreds of millions of dollars that the screenplay writer probably made less than that, a lot less, a lot less. If they're very savvy, they might've gotten a little bit of a cut of that movie, but in most cases they're getting next to nothing. In the world we're going into, if I have a great sense of story, if I know what a great movie should look like, I don't have to stop at the screenplay. I will be able to produce the entire movie probably for tens of thousands of dollars or less, including editing, sound.

1:00:54I'm not going to have to pay actors. So I would worry if I'm an actor, actually, in this world. But the creative who's at the core of the idea is going to be able to go much, much further. And then they're going to be able to self-publish it on YouTube and monetize it on YouTube. Or maybe there'll be paths. Maybe there'll be a YouTube slash Netflix-like thing where slightly vetted people can surface and put their content out. So I would be much more worried if I was the production houses. You know, the production houses probably are going to invest in it. They're going to figure out ways to do editing much cheaper.

1:01:27They're going to figure out ways, you know, I mean, they are going to be able to write screenplays. But a commodity screenplay is very different than a great screenplay. and I think for society it's going to be good because think about how many hundred million dollar duds there are how many bad movies frankly most hundred million dollar movies are bad what a waste of society's resources now we're going to have a bunch more of movies that probably cost ten or a hundred thousand dollars to make and they're going to have a higher percentage of bad ones because it's going to be like YouTube but we're also going to have a higher total quality of good ones And so it's interesting, you know, because Tyler Perry, for instance, he shut down plans to build a hundred million dollar studio space because he says AI is going to do this.

1:02:16So he, you know, he's a big movie creator. He has the same view as you. And I think I agree, but it's interesting to see how nervous the writers are. Maybe it's just like an insecurity of writers because they haven't been historically making that money. They've been kind of kept down in the system. So they just assume they're going to be kept down again. But you're right, it is going to give them power. But it's interesting, from your book, there's various kind of groups of people that have these fears. So there's creatives have these fears. There are teachers that have this existential fear. There's professors who worry the students are cheating.

1:02:52An interesting chapter is the one on parenting. Do you see parents being afraid AI could take their place? I don't think any parents, well, hopefully parents aren't afraid of AI taking their job as a parent. I think it's, you know, most of the fears of a parent are we see directly our kids getting addicted to devices. We either see directly in our own families or schools or we read about how things like social media and cell phones are affecting mental health of especially young people, teenagers, probably disproportionately young girls. And that's scary to parents. And so when we see a new technology that's as powerful as AI, okay, is this going to make everything worse?

1:03:36Not really having a clear idea of how it will make it worse. My hope, once again, is I actually think used well, I'm sure there's going to be use cases of AI that are not great. That can almost amplify a lot of the bad things that we already see on the internet. Whether it's marketing to you, whether it's forms of social media that make you feel insecure. or you have permanent fear of missing out or whatever's going on or hurt your body image, whatever. But I think there's going to be use cases. These are the ones we're going to focus on. Hopefully others focus on it as well, where the AI, beyond just helping you academically, it can help you even handle the world that you're dealing with.

1:04:20I write in the book about the AI acting as a guardian angel. Right now, when you surf the internet and our children surf the internet, they don't realize it, but they're already doing battle and they don't even know they're in a battle with AIs where these AIs are figuring out the next thing to show on their social media feed or the search results or the next ad. And their objective function is what's going to get you to click on that ad or what's going to make you watch longer. And unfortunately, it seems like triggering content, polarizing content, content that makes you feel bad about yourself is the stuff that makes you sit on it longer.

1:04:57and you know any of us who have fallen into it we've all spent some time on the internet and clicked on an ad or spent more time on social media and we never feel good about it when we're done right we all we all feel like we kind of wait is that true i feel like that's a little bit of a cliche like sometimes i surf the internet and i like i i enjoy the experience i i i for me it's true. I, I, there, there, there's definitely times where I, I, I, my friend sends me a fun video or a meme, or I fall into it a little bit on YouTube or on TikTok and I get a good giggle. And that was exactly what I needed that day.

1:05:35But more often than not, you know, I, I, we have fully developed frontal lobes and we can regulate ourselves after 10 or 15 minutes of that. But, you know, I'll say even TikTok, which I've shut down my account and I had a, I had a reason to be on it. We had a social media following. I just took it off because I found even myself, I just went to TikTok to like, you know, announce a new feature on Khan Academy. And then I end up just swiping and swiping and then 10, 15 minutes go by. I'm like, I don't feel good about what I'm doing with my day. And I stop. I was able to stop at 15 minutes. There's a lot of young people who aren't stopping at all.

1:06:09And they're going two, three, four hours. And look, it's great if someone has the self-regulation to be able to stop after 15, 20 minutes, that's probably healthy fun as long as it's not making them feel bad about themselves. But I am imagining a world now, and we're working on this, where an AI can act on your behalf or act on your parents or your teacher's behalf where, yeah, okay, I'll let my daughter use a cell phone, but I want to see a world where there is an AI agent on that phone that I am in conversation with. And I always tell them like, look, I'm cool with her spending some time on it, But, you know, let's be careful about her seeing things that hurt her body image or things that make her feel like she's missing out on things or just put her into this polarizing haze or whatever it might be.

1:06:57And then the AI can make sense of what she's surfing and looking at and say, hey, we just spent the last 10 minutes looking at, you know, pictures of that friend's birthday party. And when you were out of town, maybe we want to go do something else. or if it can see patterns in what the daughter is doing, tell the parents, hey, did you know your child's spending an awful lot of time doing X, Y, or Z? One, that might just be something to police a little bit more or it might be an opportunity to have a connection. Like, did you know that they're really interested in crocheting? Maybe you should take it up.

1:07:30Maybe you should have a conversation with them. This is something to talk about at the dinner table. So I'm hoping that the AI can act much more as a, used well, a little bit of a guardian angel, and it can help parents, it can support parents engage with their children more. I hope that, you know, right now we have devices like Siri and Alexa and Google Home. And right now they're kind of like, you know, what time is it? Put a timer on, you know, what's the weather today? You know, how many people live in Ukraine? That's the type of things we use it for. I would love a time, which I think is going to happen in the next year or two, where I'm, you know, sometimes we're having dinner together.

1:08:09I want to, and we're having a conversation, but I know that there's more that my kids have going on in their life that I'm not getting out of them. When I just ask them, how was your day? How was school? What's, you know, how's history class going? And I could imagine a world where an AI says, I could say, hey, Alexa, can you moderate a conversation, a fun conversation between us as kind of an icebreaker? And it does. And it says, all right, we're going to go around and everyone's going to say the best part of their day. And then when my nine-year-old says, oh, my best part of the day was getting to hang out with my friend who was out sick, then they can say, well, why was that?

1:08:47But once again, it's not squeezing out the parent. It's creating a context. As a parent, I crave more context where I can engage with my children in a more meaningful way, as opposed to just being transactional. Put that iPad away, go to bed, eat your food. Hey, it's time to get up. Hey, we're late for the bus. We all have to do that as parents, but it squeezes out the like, What do you think is the meaning of life? Or what does it mean to be a great friend? Or how can we all be better family members? If we could have an expert facilitator all the time, that'd be pretty incredible. Do you think, are you going to get a wearable AI, like this humane AI pin or rabbit?

1:09:27Or maybe make Conmingo a wearable? Yeah, I have no imminent plans to do that. Because I'm also someone who really likes going off the grid, so to speak, and being completely unplugged in certain cases. But if there are devices that come out that I can legitimately believe will enhance my life in some way, I'd be open to it. And as long as they have the data privacy and all of the right safeguards in place. now i'm wondering what one big question out there is does ai somehow plateau like is there only so much it could learn and simulate you know human conversation so you mentioned how like gpt5 is going to have a trillion parameters as opposed to gpt4 which just had 175 billion as opposed to gpt1 which had i don't know a million or whatever it was at a trillion parameters or 10 trillion parameters.

1:10:28What can AI do that prior versions couldn't do? None of us know for sure. And I think it is a philosophical debate about if an AI is trained on human created content for the most part, can it transcend human intelligence in certain ways? I think the answer is in certain ways it will be able to, because as these, as a parameters increase, it'll be able to find patterns that are not maybe explicitly obvious to us and maybe leverage those patterns to create things. But will it be like, you know, a truly transcendent intelligence? I don't know. I think one of the interesting things that are going to happen is they're going to be collecting more and more inputs.

1:11:15Right now, it's pretty much human-created, written text for the most part, and now images, videos, sound, files. I think when the AIs get more sensory perception, once you have wearables, once they have cameras, I know this is a little bit dystopian, so we have to think about how this gets provisioned. It's going to get more training data. I also think robotics is interesting because when we learn, we don't just observe our environment. we play with the environment. I mean, we literally, that's what, play is literally learning. A kid pokes something, throws it, you know, shakes it and figures out what, how they're investigating the world.

1:11:56And I can imagine once you pair AI with robotics and then it can actively investigate the world, it will get even more input on, and maybe even, and look, it's going to be able to have sensory input that we don't have. It will be able to see the entire electromagnetic spectrum, not just that it's going to be able to hear every frequency of sound. It's going to be able to detect molecules that we can't at least consciously detect. So it's an interesting philosophical debate. But once you start talking about orders of magnitude, more parameters than we have synapses in the human brain, and it's tireless and it has access to all of this information and it can process it faster than we can, it is interesting.

1:12:51And I know everything, even me talking out loud, I've kind of scared myself a bit. But, you know, once again, it's all about intent and how we use it and can we use it for good purposes because there can be a lot of good purposes here. I mean, to some extent, let's say a good purpose is education. I do, and there's evidence that I think with a general AI that basically exceeds our current educational efforts, kids are going to get smarter. Meaning they're just going to learn faster. They're going to retain more. They're going to get that feedback loop much more quickly. And so they're going to be able to advance more quickly than our generation was or the generations in between.

1:13:40I mean, an example that you can see in a specific domain, it's not a general domain, but look, computers have been better than humans at chess since 1997. And computers have evolved incredibly since then. So it's been over 25 years or 27 years of computers improving since they were already better than humans. and kids use computers to learn chess with some coaching, but 99 % is they're playing the computer and getting immediate feedback. And then there might be some coaching and kids are so much better now than they were back in 1997. And their rate of improvement is so much faster. I mean, a 17 year old just the other day became the challenger to the world championship that never would have happened 25 years ago.

1:14:34So it's incredible how in that specific domain and other domains like that, kids have advanced and learned and retained so much faster than kids back then. And I wonder if the same thing is going to happen in general education. I obviously hope so. It seems like it definitely will. Like our kids will be smarter. And I typically traffic in optimistic scenarios. And so I hope what you're saying is right. There is a more dystopian, I don't know if it's dystopian, but a less optimistic scenario where you're going to have a segment of students who take these tools and run with it and will do exactly what you said.

1:15:15They're going to get to heights that we never thought was possible at very young ages. And look, that by itself is not in any way a negative thing. That is a positive thing. these are going to be the young people who cure diseases for us, who start the next great companies, who write the next great novel, who produce the next great movies, all of the above, push AI forward or make sure that AI is used in a safe way. But I do, I think it's how do we make sure that the other, whatever it is, 60, 70 % of students stay engaged and engaged enough to not

1:15:56be thrown behind by all of this. Because if they are, it's not good for them as human beings, and it's not good for society. And so that's where a lot of our, what we're already seeing with Conmigo in the classroom is there is a segment of student who just runs with it, and they are off to the races. And once again, no one should hold them back. I mean, sometimes the school system's temptation is to hold those students back. That is not a good idea. We want these students to move as fast as they can. But how do we make sure that everyone has a decent chance of being able to participate without holding people back is where a lot of our work is.

1:16:33Well, and you know, you've been involved in, obviously you've been involved in education for almost 20 years. Khan Academy has been around almost that long. Why do we have the old system at all? Why do we have credentialing at all where, oh, in order to get a job in a high profile place, I need a degree from Harvard. Why can't I just go to Khan Academy for a year or two, learn what I need to learn and get a job at Goldman Sachs or Silicon Valley or Hollywood or whatever? Like, what is the real role now of higher education? Why can't I just use online education for that? I agree with you 100%. And you're starting to see aspects of that exist.

1:17:14You're definitely seeing that in fields like software engineering, where, you know, the same employers that 20 or 30 years ago might have said, okay, we're only going to hire from MIT and Caltech and Stanford. They're now saying, hey, anyone can take this boot camp or take this assessment. And if you pass it, we're going to interview you the same as we would interview a, you know, 4.0 grad from Stanford. So that's already happening in certain fields. I think what you're going to see, and this is a passion of mine, is I do want to create competency-based credentials that have the same or greater prestige as going to Harvard or going to Oxford.

1:17:55So much so that even if you go to Oxford or Harvard, you'll still want to get these things to show that you actually learned some very useful, very useful skills. So systems change is harder than technological change for a whole series of reasons. But this is something that I, you know, and hopefully I have many decades left on this planet, but I'm hoping in the next 10 or 20 years, hopefully closer to five or 10 years, you're going to hear some things, even from us, about ways that you can get high school, college credit, potentially even job opportunities via a pretty streamlined route. Once again, it doesn't mean that it's an either or.

1:18:32You might want to do both. But if you don't have access to Harvard, which very, you know, I was telling some very senior people at Harvard that right now there's this false tension where they feel, you know, everyone writes about admissions and it feels like a false tension between equity and merit where there's like, oh, affirmative action that gets ruled down by the Supreme Court. So how are we going to make sure we have equity and that we have right, you know, good representation if we have to go more based purely on test scores or whatever else. And what I've told them is like, you know, it's a false tension because what you've held constant is capacity.

1:19:09Like you only admit whatever 2 ,000 students every year, even though there's probably 100 ,000 qualified students every year. If you could serve all of the 100 ,000 qualified students, you wouldn't have to have these weird, bizarre discussions about admissions and equity and all of this that people are having, you would just be able to, if you're, if you can handle the work, we are admitting you and you will get a credential that can get, that can open up the world to you. So yes, I hope that we can, we can create some pathways like that. I think though, part of that tension is, is the scarcity principle, which is that Harvard prices their scarcity.

1:19:48And so they make money from their scarcity. And Khan Academy, or let's say a commercial equivalent, makes money from scalability as opposed to scarcity. And Harvard doesn't want to have 100 ,000 students. They want to accept only 2 ,000 because they could charge 300 ,000 a semester and they'd still fill up. And I think that's the tension really. Yeah. And obviously our motivation is, our costs go up as we scale. So our motivation is more of just like, well, how do we maximize impact? But you're right about there's always going to be a notion of scarcity and the reality of the Harvards of the world.

1:20:26And so much of the education debate focuses on, let's call it about 30 universities, even though it's less than 1 % of students go to those 30 universities. But those 30 universities are relevant because they do over-influence society. And so I think there are, it's not going to go away anytime soon that there's a certain, you know, the U.S. doesn't have a caste system in the traditional sense, but our elite institutions are the best proxy for them. Where, oh, oh, you're Harvard class of whatever, were you there? Oh, did you have this professor? Oh, wow. Like we're part of the same, part of the same club.

1:21:06I don't think that's going to go away. what I, and I think there's other positives of not just Harvard or Stanford, but college in general of being there in person, forming bonds. I met my wife and many of my best friends in college and we have our best memories there. So I think we can always optimize for some of those in-person experiences, but I want to create also alternative pathways. So that's not the only way that you can get into an upper middle class or, you know, promising career. On that optimistic note, and I really hope you're right. I've been writing about this also for a long time.

1:21:40And I'm really looking forward to the utopian world that you have been outlining and creating and explaining, particularly in your book. I'm going to say the title again, because I always forget things really quickly. Brave New Words, How AI Will Revolutionize Education and why that's a good thing. And I think what you've explained today has solved a lot of the, also the fears that many people have about AI, but Salman Khan, creator of the Khan Academy. Thanks once again for coming on the show. Thanks for having me, James.